1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
use num_traits::Bounded;
#[cfg(feature = "dtype-struct")]
use polars_core::prelude::sort::arg_sort_multiple::_get_rows_encoded_ca;
use polars_core::prelude::*;
use polars_core::series::IsSorted;
use polars_core::with_match_physical_numeric_polars_type;
use polars_utils::total_ord::TotalOrd;

use crate::series::ops::SeriesSealed;

pub trait SeriesMethods: SeriesSealed {
    /// Create a [`DataFrame`] with the unique `values` of this [`Series`] and a column `"counts"`
    /// with dtype [`IdxType`]
    fn value_counts(&self, sort: bool, parallel: bool) -> PolarsResult<DataFrame> {
        let s = self.as_series();
        polars_ensure!(
            s.name() != "count",
            Duplicate: "using `value_counts` on a column named 'count' would lead to duplicate column names"
        );
        // we need to sort here as well in case of `maintain_order` because duplicates behavior is undefined
        let groups = s.group_tuples(parallel, sort)?;
        let values = unsafe { s.agg_first(&groups) };
        let counts = groups.group_count().with_name("count");
        let cols = vec![values, counts.into_series()];
        let df = unsafe { DataFrame::new_no_checks(cols) };
        if sort {
            df.sort(
                ["count"],
                SortMultipleOptions::default()
                    .with_order_descending(true)
                    .with_multithreaded(parallel),
            )
        } else {
            Ok(df)
        }
    }

    #[cfg(feature = "hash")]
    fn hash(&self, build_hasher: ahash::RandomState) -> UInt64Chunked {
        let s = self.as_series().to_physical_repr();
        match s.dtype() {
            DataType::List(_) => {
                let mut ca = s.list().unwrap().clone();
                crate::chunked_array::hash::hash(&mut ca, build_hasher)
            },
            _ => {
                let mut h = vec![];
                s.0.vec_hash(build_hasher, &mut h).unwrap();
                UInt64Chunked::from_vec(s.name(), h)
            },
        }
    }

    /// Checks if a [`Series`] is sorted. Tries to fail fast.
    fn is_sorted(&self, options: SortOptions) -> PolarsResult<bool> {
        let s = self.as_series();
        let null_count = s.null_count();

        // fast paths
        if (options.descending
            && (options.nulls_last || null_count == 0)
            && matches!(s.is_sorted_flag(), IsSorted::Descending))
            || (!options.descending
                && (!options.nulls_last || null_count == 0)
                && matches!(s.is_sorted_flag(), IsSorted::Ascending))
        {
            return Ok(true);
        }

        // for struct types we row-encode and recurse
        #[cfg(feature = "dtype-struct")]
        if matches!(s.dtype(), DataType::Struct(_)) {
            let encoded =
                _get_rows_encoded_ca("", &[s.clone()], &[options.descending], options.nulls_last)?;
            return encoded.into_series().is_sorted(options);
        }

        let s_len = s.len();
        if null_count == s_len {
            // All nulls is all equal
            return Ok(true);
        }
        // Check if nulls are in the right location.
        if null_count > 0 {
            // The slice triggers a fast null count
            if options.nulls_last {
                if s.slice((s_len - null_count) as i64, null_count)
                    .null_count()
                    != null_count
                {
                    return Ok(false);
                }
            } else if s.slice(0, null_count).null_count() != null_count {
                return Ok(false);
            }
        }

        if s.dtype().is_numeric() {
            with_match_physical_numeric_polars_type!(s.dtype(), |$T| {
                let ca: &ChunkedArray<$T> = s.as_ref().as_ref().as_ref();
                return Ok(is_sorted_ca_num::<$T>(ca, options))
            })
        }

        let cmp_len = s_len - null_count - 1; // Number of comparisons we might have to do
                                              // TODO! Change this, allocation of a full boolean series is too expensive and doesn't fail fast.
                                              // Compare adjacent elements with no-copy slices that don't include any nulls
        let offset = !options.nulls_last as i64 * null_count as i64;
        let (s1, s2) = (s.slice(offset, cmp_len), s.slice(offset + 1, cmp_len));
        let cmp_op = if options.descending {
            Series::gt_eq
        } else {
            Series::lt_eq
        };
        Ok(cmp_op(&s1, &s2)?.all())
    }
}

fn check_cmp<T: NumericNative, Cmp: Fn(&T, &T) -> bool>(
    vals: &[T],
    f: Cmp,
    previous: &mut T,
) -> bool {
    let mut sorted = true;

    // Outer loop so we can fail fast
    // Inner loop will auto vectorize
    for c in vals.chunks(1024) {
        // don't early stop or branch
        // so it autovectorizes
        for v in c {
            sorted &= f(previous, v);
            *previous = *v;
        }
        if !sorted {
            return false;
        }
    }
    sorted
}

// Assumes nulls last/first is already checked.
fn is_sorted_ca_num<T: PolarsNumericType>(ca: &ChunkedArray<T>, options: SortOptions) -> bool {
    if let Ok(vals) = ca.cont_slice() {
        let mut previous = vals[0];
        return if options.descending {
            check_cmp(vals, |prev, c| prev.tot_ge(c), &mut previous)
        } else {
            check_cmp(vals, |prev, c| prev.tot_le(c), &mut previous)
        };
    };

    if ca.null_count() == 0 {
        let mut previous = if options.descending {
            T::Native::max_value()
        } else {
            T::Native::min_value()
        };
        for arr in ca.downcast_iter() {
            let vals = arr.values();

            let sorted = if options.descending {
                check_cmp(vals, |prev, c| prev.tot_ge(c), &mut previous)
            } else {
                check_cmp(vals, |prev, c| prev.tot_le(c), &mut previous)
            };
            if !sorted {
                return false;
            }
        }
        return true;
    };

    // Slice off nulls and recurse.
    let null_count = ca.null_count();
    if options.nulls_last {
        let ca = ca.slice(0, ca.len() - null_count);
        is_sorted_ca_num(&ca, options)
    } else {
        let ca = ca.slice(null_count as i64, ca.len() - null_count);
        is_sorted_ca_num(&ca, options)
    }
}

impl SeriesMethods for Series {}